Deep Learning-Based Survival Analysis for High-Dimensional Survival Data

With the development of high-throughput technologies, more and high-dimensional or ultra-high-dimensional genomic data are being generated. Therefore, effectively analyzing such has become a significant challenge. Machine learning (ML) algorithms have been widely applied for modeling nonlinear complicated interactions in variety practical fields as survival data. Recently, multilayer deep neural network (DNN) models made remarkable achievements. Thus, Cox-based DNN prediction model (DNNSurv model), which was built with Keras TensorFlow, developed. However, its results were only evaluated on datasets large sample sizes. In this paper, we performance DNNSurv using compared it three popular ML (i.e., random forest LASSO Ridge models). For purpose, also present optimal setting several hyperparameters, including selection tuning parameter. The proposed method demonstrated via analysis that performed well overall models, terms main evaluation measures concordance index, time-dependent Brier score, AUC) performance.